Synthetic Trust: Exploiting Biases at Scale
summary
AI voice cloning already produces real financial harm: a 7,000-person, 11-country survey finds 1 in 4 people affected (India 50%), average losses above $11,000, and fraud volumes spike to 10.2 million attacks after ChatGPT. Attackers use scraped voicemail or social audio plus open models (Llama 2, Megabyte) to create real-time voice clones, replayed asynchronously to extort thousands; humans detect fakes only ~57% of the time while t-SNE shows synthetic audio tightly clustered. A prototype synthetic-fraud adapter alerts landline users in real time; deep-fake CAPTCHA research reports >90% accuracy for anti-clone challenges.
AI-written summary · about
chapters (7)
- 0:00Synthetic trust overview
- 7:08Synthetic audio rise and risks
- 13:48Real cases of audio fraud
- 17:00Exploiting voice biases
- 26:10Protecting vulnerable groups
- 31:56Detecting synthetic audio
- 36:59AI fraud implications
related talks
36mFabricated Realities: AI Personas and Simulacra and the Threats They Pose
36mDemocratized Deception: How Threat Actors Are Weaponizing AI to Scale Cybercrime
39m Put Words In My Mouth
20mSecurity in the Age of AI: Response and Strategy
47mFighting AI with AI
30mAI-Generated Synthetic Media, aka Deepfakes and its impact
45mFUD vs. Reality: Bracing for an AI-Powered Threat Landscape
40mBanking Cybersecurity: Battling DeepFakes & AI-powered Scammers